Data Integration for Satellite Precipitation Estimation Using Deep Neural Networks

Abstract:

Recent developments in meteorological satellite remote sensing, along with advancements in machine learning techniques and computational power, open great opportunities to integrate massive amounts of real-time observations to characterize both vertical and horizontal structure of clouds and their precipitation potentials. Our specific objective is to adapt an “end-to-end” deep learning architecture to improve precipitation estimation using multi-satellite remotely sensed information. In this work, a deep convolutional neural network (CNN) is applied to estimate precipitation using a synergy of infrared (IR) data from geosynchronous-Earth-orbit platforms and active and passive microwave (PMW) observations from Global Precipitation Measurement (GPM) satellite. The GPM Dual-frequency Precipitation Radar (DPR) and the multifrequency passive microwave radiometer (GMI) provide unprecedented three-dimensional spaceborne information of the structure of precipitation systems. We combine brightness temperatures from GPM GMI, radar reflectivity measurements from both Ka- and Ku-bands, and Geostationary IR cloud-top brightness temperatures to the gauge-adjusted Multi-Radar Multi-Sensor (MRMS) as reference observations. The matching is performed through an automatic big data preprocessing pipeline at the time and location of each GPM orbits over CONUS. About 57,000 collocated samples between multiple sources within ±30 mins time difference (from January 2017 to December 2018) are obtained to train and test the deep CNN precipitation model. Overall, our experiments demonstrate that the multi-sensor data integration scheme together with the optimized deep CNN algorithm has a better performance compared to single-source precipitation retrievals (e.g. IR- and PMW-based). The results show the efficiency of the proposed model in surface precipitation estimation while it mitigates the underestimations of heavy precipitation rates and precipitation from warm clouds.